HELM VHELM - Blink Perception - Relative Depth: leaderboard

Metric: Prefix Quasi-Exact Match (%). Source: crfm.stanford.edu. 48 models tracked.

Top models

#ModelScore
1Llama 4 Scout Instruct96.77
2Llama 4 Maverick Instruct FP891.94
3Gemini 2.5 Pro (Preview 03-25)90.32
4Claude 3 Opus (20240229)89.52
5GPT-4.588.71
6GPT-4.1 (2025-04-14)87.1
7Gemini 2.0 Flash Lite87.1
8O3 (2025-04-16)86.29
9Gemini 1.5 Pro (002)84.68
10Gemini 1.5 Flash (002)83.87
11Gemini 2.0 Pro (Preview 02-05)83.87
12Gemini 2.0 Flash83.06
13O4 Mini (2025-04-16)83.06
14Claude 3 Sonnet (20240229)82.26
15GPT-4.1 Mini81.45

Interactive version: theaggregate.ai/benchmark?slug=helm-vhelm-blink-perception-relative-depth · How It Works · Data refreshed daily, snapshot 2026-09-19.